High-Throughput Mapping of Long-Range Neuronal Projection Using In Situ Sequencing

Abstract

Understanding neural circuits requires deciphering interactions among myriad cell types defined by spatial organization, connectivity, gene expression, and other properties. Resolving these cell types requires both single-neuron resolution and high throughput, a challenging combination with conventional methods. Here, we introduce barcoded anatomy resolved by sequencing (BARseq), a multiplexed method based on RNA barcoding for mapping projections of thousands of spatially resolved neurons in a single brain and relating those projections to other properties such as gene or Cre expression. Mapping the projections to 11 areas of 3,579 neurons in mouse auditory cortex using BARseq confirmed the laminar organization of the three top classes (intratelencephalic [IT], pyramidal tract-like [PT-like], and corticothalamic [CT]) of projection neurons. In depth analysis uncovered a projection type restricted almost exclusively to transcriptionally defined subtypes of IT neurons. By bridging anatomical and transcriptomic approaches at cellular resolution with high throughput, BARseq can potentially uncover the organizing principles underlying the structure and formation of neural circuits.

Keywords: auditory cortex; cellular barcoding; high throughput; in situ sequencing; projection mapping.

Fig. 1.

Multiplexed projection mapping using in situ sequencing. (A) Workflow of MAPseq (left) and BARseq (right). (B) NGS vs. in situ sequencing. (C) Comparison of in situ sequencing and hybridization techniques. These techniques may use multiple rounds of hybridization to probe and read out multiple mRNAs (a)(b), use sequencing to multiplex read out of hybridization signals (c)(d), or copy target sequences from the mRNA into the rolonies to allow true sequencing (e)(f). (D) Representative images of barcode rolonies generated in primary hippocampal neuronal culture coexpressing GFP. All GFP positive neurons were filled with barcode amplicons, indicating efficient barcode amplification in neuronal somata. (E) Images of the first four sequencing cycles of the same neurons shown in (D). The bases corresponding to the four colors and the sequences of the three neurons circled in (E) are indicated to the right. In all images, scale bars = 50 μm.

Fig. 2.

Validation of BARseq. (A) Representative low-magnification images of GFP (left) co-expressed with barcodes and rolonies (right) in a brain slice. GFP intensity does not correlate perfectly with rolony intensity due to differences in protein and RNA expression. Insets: Negative control images of GFP and rolonies of a non-barcoded brain slice taken with the same exposure settings. No GFP or rolonies are visible in these images. Scale bars = 50 μm. (B) Representative high-resolution sequencing image of a barcoded brain slice. Scale bar = 50 μm. (C) Low-resolution images of the indicated cycles of barcode sequencing in a brain slice. The sequences of the three cells are indicated below. Scale bars = 100 μm. (D) The quality of the base calls on the barcoded brain slice. (E) Histogram of the number of mismatches between the in situ reads and their closest matches from in vitro reads (in situ) and the number of mismatches between random sequences and their closest matches from in vitro reads (Random). (F) An example barcode read in situ and its closest match in vitro, and a random sequence and its closest match in vitro. Red indicates mismatches. (G) A brain was injected with CTB in the contralateral auditory cortex and barcoded in the ipsilateral auditory cortex. BARseq results of the barcoded neurons were then compared to retrograde labeling by CTB. (H) A representative image of a brain slice double labeled with barcodes (cyan) and CTB (magenta) from the contralateral auditory cortex. Dashed lines indicate the top and the bottom of the cortex. Scale bar = 100 μm. A magnified view of the squared area is shown in the inset. The arrow in the inset indicates a GFP+ CTB+ double-labeled neuron. (I) Venn diagram showing the number of GFP expressing neurons labeled with (magenta) or without (white) CTB and/or neurons found to project contralaterally using BARseq (cyan). See also Fig. S1 and Table S1.

Fig. 3.

Correlating gene expression and projections using BARseq. (A) Barcode sequencing in Cre-labeled animals. Top: images of Fezf2+ tdTomato expressing cells (left), GFP expressing barcoded cells (middle), and merged image of the two (right). Scale bar = 100 μm. Bottom: sequencing images of the boxed area in the merged image. Arrows indicate tdTomato-expressing barcoded neurons. (B) Projection patterns of all neurons (left) and Fezf2+ neurons (right). Rows indicate neurons and columns indicate projection areas. (C) The cortical depth distribution of barcoded Fezf2+ somata color-coded by projections. (D) Representative images of FISH (left) and sequencing images (right) of the same sample. (E) The expression profile of Slc17a7 and Gad2 and barcode sequences of the neurons boxed in (D). Scale bars = 50 μm. (F) The number of neurons with or without projections (Proj) in cells that did or did not express Slc17a7 (S) and/or Gad2 (G). (G) Mean log normalized expression of each gene averaged over all barcoded (x-axis) and non-barcoded (y-axis) neurons. The gene expression is regressed with a Poisson model to remove the effect of both the percentage of mitochondrial genes and endogenous UMI counts. See also Fig. S2 and Table S2.

Fig. 4.

Mapping projections of the auditory cortex using BARseq. (A) Histogram of the minimal pairwise hamming distance of the first 15 bases of barcodes recovered from brain XC9. (B) Single-cell projection patterns sorted by clusters. (C) Conventional bulk GFP tracing intensities (x-axis) were plotted against the bulk projection strength obtained from BARseq (y-axis). Error bars indicate SEM. N = 5 for GFP tracing and N = 3 for BARseq. Pearson correlation coefficient r = 0.94, p < 0.0001. (D) The distribution of projection intensity in each projection area. The y-axis indicates the logarithms of raw barcode counts in each area, and the x-axis indicates the number of cells. (E) The numbers of binarized projection patterns (y-axis) after filtering for primary projection strength (x-axis, blue line) or after random subsampling to the same sample size (red line). Black lines and error bars indicate 95% confidence interval for subsampling. (F) Histogram of the number of projections per neuron. (G) The fractions of multi-projection neurons (y-axis) are plotted against the ratio between the secondary and primary projections (x-axis). Blue line indicates actual distribution and red line indicates fitting with a log normal distribution. See also Fig. S3 and Table S3.

Fig. 5.

The laminar organization of projection neurons in the auditory cortex. (A) The mean projection patterns of clusters corresponding to the indicated major classes of neurons. Line thickness indicates projection strength normalized to the strongest projection for that class. Blue arrows indicate projections to contralateral brain areas and black arrows indicate projections to ipsilateral brain areas. (B) The sequenced projection neurons from a brain (XC9) are color-coded by class identities and plotted at their locations in the cortex. The top and bottom of the cortex are indicated by the red and blue dashed lines, respectively. The laminae and their boundaries are marked. Scale bar = 100 μm. Inset: histograms of the laminar depths of each class of projection neurons in the pooled BARseq dataset. (C) Hierarchical clustering of single-cell projection data. Top: dendrogram of the hierarchical structure of the clusters. Middle: the mean projection patterns of the corresponding leaf clusters. Bottom: The laminar distribution of the corresponding leaf clusters. Individual neurons are superimposed on top of the distribution plots (light grey). Neurons whose cluster identity were less confident were marked in gray. The number of cells that belong to each leaf cluster is indicated below. Neurons of subcluster 25 were likely misidentified PT-l neurons (see STAR Methods). (D) t-SNE plot of the projection neurons. The neurons are color-coded by their first level subcluster identities post-hoc. (E) The normalized entropy of nodes/leaves (y-axis) in the indicated clustering hierarchy (x-axis). Grey bars indicate mean ± stdev of all nodes/leaves of a specific hierarchy. See also Fig. S4–S6.

Fig. 6.

Subtypes of IT neurons defined by gene expression in the auditory cortex. (A) Histograms of the log normalized expression of the indicated marker genes in the indicated clusters obtained from single-cell RNAseq in the auditory cortex. The dendrograms show distances of mean gene expression among transcriptomic clusters (left) and distances of mean projection pattern (right) obtained through BARseq and FISH. (B) t-SNE plot of the gene expression of neurons color-coded by cluster identity as in (A). (C) MetaNeighbor comparison of neuronal clusters obtained in the auditory cortex to those in the visual cortex from Tasic et al. (2018). (D) Projections (left) and the expression of genes (right) of neurons obtained using combination of BARseq and FISH are shown on a log scale. Projection areas are the same as in Fig. 4B, except that each cortical area is divided into upper (u) and lower (l) layers. (E) Distributions of laminar positions of neurons. Individual neurons (red) are superimposed on the smoothed distribution (black). See also Fig. S7.

Fig. 7.

Projections across IT subtypes defined by gene expression. (A) The fraction of ITi-Ctx neurons in each indicated IT subtype defined by gene expression. The numbers of ITi-Ctx neurons and the total numbers of each subtype are indicated. Inset: the mean projection pattern of ITi-Ctx neurons. (B) Histograms of the layer bias of contralateral projections of neurons in each IT subtype. The histograms are normalized so that the maximum value for a bin is 1 for each subtype. (C) The log normalized barcode count of projections to the contralateral auditory cortex (x-axis) is plotted against that of projections to the ipsilateral visual and somatosensory cortex (y-axis) for each neuron. (D) The fractions of neurons of IT subtypes defined by gene expression that belong to each indicated IT projection leaf subcluster. All bars belonging to a transcriptionally defined subtype sum to 1 across the whole plot. (E) The number of neurons with (AudC+) or without (AudC−) projections to the contralateral auditory cortex in each IT subtype defined by gene expression. Neurons in each IT subtype are further divided into those expressing Cdh13 and those that do not. *p < 0.005 using Fisher’s exact test after Bonferroni correction. (F) Two example projection leaf subclusters that were shared across all four IT subtypes defined by gene expression. Projection diagrams indicate example neurons. The numbers of neurons of a transcriptionally defined subtype and of those belonging to a projection subcluster are indicated below. (G) The minimum distance in projections from a neuron to any neuron within the same subtype defined by gene expression (x-axis) or in a different subtype (y-axis). See also Fig. S7.